Missed optimization opportunity when trivially moving struct by moving fields
Nobody has claimed this yet.
- Dominant language
- Rust
- Stars
- 119k
- Forks
- 16.1k
- PR merge metrics
- PR metrics pending
Description
Example code:
#[repr(C)]
pub struct ThreeFields<'a> {
a: &'a [u32],
b: &'a [u32],
c: &'a [u32],
}
#[inline(always)]
pub fn should_be_no_op(val: ThreeFields) -> ThreeFields {
ThreeFields {
a: val.a,
b: val.b,
c: val.c,
}
}
pub fn sum_slices_1(val: ThreeFields) -> u32 {
sum(&val)
}
pub fn sum_slices_2(val: ThreeFields) -> u32 {
let val = should_be_no_op(val);
sum(&val)
}
#[inline(never)]
pub fn sum(val: &ThreeFields) -> u32 {
val.a.iter().sum::<u32>() + val.b.iter().sum::<u32>() + val.c.iter().sum::<u32>()
}
In rustc 1.84 stable this generates a number of moves that I don't think need to be there, especially when inlining:
example::sum_slices_1::hc2a0527df8a4985d:
jmp qword ptr [rip + example::sum::h228167780f08fdbb@GOTPCREL]
example::sum_slices_2::hfa5e09ed5de6d084:
sub rsp, 56
movups xmm0, xmmword ptr [rdi]
movups xmm1, xmmword ptr [rdi + 16]
movups xmm2, xmmword ptr [rdi + 32]
movups xmmword ptr [rsp + 8], xmm0
movups xmmword ptr [rsp + 24], xmm1
movups xmmword ptr [rsp + 40], xmm2
lea rdi, [rsp + 8]
call qword ptr [rip + example::sum::h228167780f08fdbb@GOTPCREL]
add rsp, 56
ret
See https://rust.godbolt.org/z/erxfM53Kq
While this is a pretty pointless example, this comes up in situations where you might want to convert a tuple of slices into a struct of slices in order to assign names to the tuple members (I've been running into these issues working on gecs).
See also: https://github.com/rust-lang/rust/issues/107436 and https://github.com/rust-lang/rust/issues/135787
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with the example and generated assembly in the linked Godbolt reproducer, comparing sum_slices_1 with sum_slices_2. Read related issues #107436 and #135787 for context; no source file or test is named. Done means the trivially moved struct no longer produces the unnecessary moves while preserving the demonstrated behavior.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- rust
- Domain
- compilers, performance
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 42/100